drivers/net/xen-netback/xenbus.c

   1 /*
   2  * Xenbus code for netif backend
   3  *
   4  * Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
   5  * Copyright (C) 2005 XenSource Ltd
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "common.h"
  22 #include <linux/vmalloc.h>
  23 #include <linux/rtnetlink.h>
  24
  25 struct backend_info {
  26         struct xenbus_device *dev;
  27         struct xenvif *vif;
  28
  29         /* This is the state that will be reflected in xenstore when any
  30          * active hotplug script completes.
  31          */
  32         enum xenbus_state state;
  33
  34         enum xenbus_state frontend_state;
  35         struct xenbus_watch hotplug_status_watch;
  36         u8 have_hotplug_status_watch:1;
  37
  38         const char *hotplug_script;
  39 };
  40
  41 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue);
  42 static void connect(struct backend_info *be);
  43 static int read_xenbus_vif_flags(struct backend_info *be);
  44 static int backend_create_xenvif(struct backend_info *be);
  45 static void unregister_hotplug_status_watch(struct backend_info *be);
  46 static void xen_unregister_watchers(struct xenvif *vif);
  47 static void set_backend_state(struct backend_info *be,
  48                               enum xenbus_state state);
  49
  50 #ifdef CONFIG_DEBUG_FS
  51 struct dentry *xen_netback_dbg_root = NULL;
  52
  53 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  54 {
  55         struct xenvif_queue *queue = m->private;
  56         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  57         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  58         struct netdev_queue *dev_queue;
  59
  60         if (tx_ring->sring) {
  61                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  62
  63                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  64                            tx_ring->nr_ents);
  65                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  66                            sring->req_prod,
  67                            sring->req_prod - sring->rsp_prod,
  68                            tx_ring->req_cons,
  69                            tx_ring->req_cons - sring->rsp_prod,
  70                            sring->req_event,
  71                            sring->req_event - sring->rsp_prod);
  72                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  73                            sring->rsp_prod,
  74                            tx_ring->rsp_prod_pvt,
  75                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  76                            sring->rsp_event,
  77                            sring->rsp_event - sring->rsp_prod);
  78                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  79                            queue->pending_prod,
  80                            queue->pending_cons,
  81                            nr_pending_reqs(queue));
  82                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  83                            queue->dealloc_prod,
  84                            queue->dealloc_cons,
  85                            queue->dealloc_prod - queue->dealloc_cons);
  86         }
  87
  88         if (rx_ring->sring) {
  89                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  90
  91                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  92                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  93                            sring->req_prod,
  94                            sring->req_prod - sring->rsp_prod,
  95                            rx_ring->req_cons,
  96                            rx_ring->req_cons - sring->rsp_prod,
  97                            sring->req_event,
  98                            sring->req_event - sring->rsp_prod);
  99                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
 100                            sring->rsp_prod,
 101                            rx_ring->rsp_prod_pvt,
 102                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
 103                            sring->rsp_event,
 104                            sring->rsp_event - sring->rsp_prod);
 105         }
 106
 107         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
 108                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
 109                    "remaining: %lu, expires: %lu, now: %lu\n",
 110                    queue->napi.state, queue->napi.weight,
 111                    skb_queue_len(&queue->tx_queue),
 112                    timer_pending(&queue->credit_timeout),
 113                    queue->credit_bytes,
 114                    queue->credit_usec,
 115                    queue->remaining_credit,
 116                    queue->credit_timeout.expires,
 117                    jiffies);
 118
 119         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
 120
 121         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
 122                    queue->rx_queue_len, queue->rx_queue_max,
 123                    skb_queue_len(&queue->rx_queue),
 124                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
 125
 126         return 0;
 127 }
 128
 129 #define XENVIF_KICK_STR "kick"
 130 #define BUFFER_SIZE     32
 131
 132 static ssize_t
 133 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 134                      loff_t *ppos)
 135 {
 136         struct xenvif_queue *queue =
 137                 ((struct seq_file *)filp->private_data)->private;
 138         int len;
 139         char write[BUFFER_SIZE];
 140
 141         /* don't allow partial writes and check the length */
 142         if (*ppos != 0)
 143                 return 0;
 144         if (count >= sizeof(write))
 145                 return -ENOSPC;
 146
 147         len = simple_write_to_buffer(write,
 148                                      sizeof(write) - 1,
 149                                      ppos,
 150                                      buf,
 151                                      count);
 152         if (len < 0)
 153                 return len;
 154
 155         write[len] = '\0';
 156
 157         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 158                 xenvif_interrupt(0, (void *)queue);
 159         else {
 160                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 161                         queue->id);
 162                 count = -EINVAL;
 163         }
 164         return count;
 165 }
 166
 167 static int xenvif_dump_open(struct inode *inode, struct file *filp)
 168 {
 169         int ret;
 170         void *queue = NULL;
 171
 172         if (inode->i_private)
 173                 queue = inode->i_private;
 174         ret = single_open(filp, xenvif_read_io_ring, queue);
 175         filp->f_mode |= FMODE_PWRITE;
 176         return ret;
 177 }
 178
 179 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 180         .owner = THIS_MODULE,
 181         .open = xenvif_dump_open,
 182         .read = seq_read,
 183         .llseek = seq_lseek,
 184         .release = single_release,
 185         .write = xenvif_write_io_ring,
 186 };
 187
 188 static void xenvif_debugfs_addif(struct xenvif *vif)
 189 {
 190         struct dentry *pfile;
 191         int i;
 192
 193         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 194                 return;
 195
 196         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 197                                                   xen_netback_dbg_root);
 198         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 199                 for (i = 0; i < vif->num_queues; ++i) {
 200                         char filename[sizeof("io_ring_q") + 4];
 201
 202                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 203                         pfile = debugfs_create_file(filename,
 204                                                     S_IRUSR | S_IWUSR,
 205                                                     vif->xenvif_dbg_root,
 206                                                     &vif->queues[i],
 207                                                     &xenvif_dbg_io_ring_ops_fops);
 208                         if (IS_ERR_OR_NULL(pfile))
 209                                 pr_warn("Creation of io_ring file returned %ld!\n",
 210                                         PTR_ERR(pfile));
 211                 }
 212         } else
 213                 netdev_warn(vif->dev,
 214                             "Creation of vif debugfs dir returned %ld!\n",
 215                             PTR_ERR(vif->xenvif_dbg_root));
 216 }
 217
 218 static void xenvif_debugfs_delif(struct xenvif *vif)
 219 {
 220         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 221                 return;
 222
 223         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 224                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 225         vif->xenvif_dbg_root = NULL;
 226 }
 227 #endif /* CONFIG_DEBUG_FS */
 228
 229 static int netback_remove(struct xenbus_device *dev)
 230 {
 231         struct backend_info *be = dev_get_drvdata(&dev->dev);
 232
 233         set_backend_state(be, XenbusStateClosed);
 234
 235         unregister_hotplug_status_watch(be);
 236         if (be->vif) {
 237                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 238                 xen_unregister_watchers(be->vif);
 239                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 240                 xenvif_free(be->vif);
 241                 be->vif = NULL;
 242         }
 243         kfree(be->hotplug_script);
 244         kfree(be);
 245         dev_set_drvdata(&dev->dev, NULL);
 246         return 0;
 247 }
 248
 249
 250 /**
 251  * Entry point to this code when a new device is created.  Allocate the basic
 252  * structures and switch to InitWait.
 253  */
 254 static int netback_probe(struct xenbus_device *dev,
 255                          const struct xenbus_device_id *id)
 256 {
 257         const char *message;
 258         struct xenbus_transaction xbt;
 259         int err;
 260         int sg;
 261         const char *script;
 262         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 263                                           GFP_KERNEL);
 264         if (!be) {
 265                 xenbus_dev_fatal(dev, -ENOMEM,
 266                                  "allocating backend structure");
 267                 return -ENOMEM;
 268         }
 269
 270         be->dev = dev;
 271         dev_set_drvdata(&dev->dev, be);
 272
 273         sg = 1;
 274
 275         do {
 276                 err = xenbus_transaction_start(&xbt);
 277                 if (err) {
 278                         xenbus_dev_fatal(dev, err, "starting transaction");
 279                         goto fail;
 280                 }
 281
 282                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 283                 if (err) {
 284                         message = "writing feature-sg";
 285                         goto abort_transaction;
 286                 }
 287
 288                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 289                                     "%d", sg);
 290                 if (err) {
 291                         message = "writing feature-gso-tcpv4";
 292                         goto abort_transaction;
 293                 }
 294
 295                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 296                                     "%d", sg);
 297                 if (err) {
 298                         message = "writing feature-gso-tcpv6";
 299                         goto abort_transaction;
 300                 }
 301
 302                 /* We support partial checksum setup for IPv6 packets */
 303                 err = xenbus_printf(xbt, dev->nodename,
 304                                     "feature-ipv6-csum-offload",
 305                                     "%d", 1);
 306                 if (err) {
 307                         message = "writing feature-ipv6-csum-offload";
 308                         goto abort_transaction;
 309                 }
 310
 311                 /* We support rx-copy path. */
 312                 err = xenbus_printf(xbt, dev->nodename,
 313                                     "feature-rx-copy", "%d", 1);
 314                 if (err) {
 315                         message = "writing feature-rx-copy";
 316                         goto abort_transaction;
 317                 }
 318
 319                 /*
 320                  * We don't support rx-flip path (except old guests who don't
 321                  * grok this feature flag).
 322                  */
 323                 err = xenbus_printf(xbt, dev->nodename,
 324                                     "feature-rx-flip", "%d", 0);
 325                 if (err) {
 326                         message = "writing feature-rx-flip";
 327                         goto abort_transaction;
 328                 }
 329
 330                 err = xenbus_transaction_end(xbt, 0);
 331         } while (err == -EAGAIN);
 332
 333         if (err) {
 334                 xenbus_dev_fatal(dev, err, "completing transaction");
 335                 goto fail;
 336         }
 337
 338         /*
 339          * Split event channels support, this is optional so it is not
 340          * put inside the above loop.
 341          */
 342         err = xenbus_printf(XBT_NIL, dev->nodename,
 343                             "feature-split-event-channels",
 344                             "%u", separate_tx_rx_irq);
 345         if (err)
 346                 pr_debug("Error writing feature-split-event-channels\n");
 347
 348         /* Multi-queue support: This is an optional feature. */
 349         err = xenbus_printf(XBT_NIL, dev->nodename,
 350                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 351         if (err)
 352                 pr_debug("Error writing multi-queue-max-queues\n");
 353
 354         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
 355         if (IS_ERR(script)) {
 356                 err = PTR_ERR(script);
 357                 xenbus_dev_fatal(dev, err, "reading script");
 358                 goto fail;
 359         }
 360
 361         be->hotplug_script = script;
 362
 363         err = xenbus_switch_state(dev, XenbusStateInitWait);
 364         if (err)
 365                 goto fail;
 366
 367         be->state = XenbusStateInitWait;
 368
 369         /* This kicks hotplug scripts, so do it immediately. */
 370         err = backend_create_xenvif(be);
 371         if (err)
 372                 goto fail;
 373
 374         return 0;
 375
 376 abort_transaction:
 377         xenbus_transaction_end(xbt, 1);
 378         xenbus_dev_fatal(dev, err, "%s", message);
 379 fail:
 380         pr_debug("failed\n");
 381         netback_remove(dev);
 382         return err;
 383 }
 384
 385
 386 /*
 387  * Handle the creation of the hotplug script environment.  We add the script
 388  * and vif variables to the environment, for the benefit of the vif-* hotplug
 389  * scripts.
 390  */
 391 static int netback_uevent(struct xenbus_device *xdev,
 392                           struct kobj_uevent_env *env)
 393 {
 394         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 395
 396         if (!be)
 397                 return 0;
 398
 399         if (add_uevent_var(env, "script=%s", be->hotplug_script))
 400                 return -ENOMEM;
 401
 402         if (!be->vif)
 403                 return 0;
 404
 405         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 406 }
 407
 408
 409 static int backend_create_xenvif(struct backend_info *be)
 410 {
 411         int err;
 412         long handle;
 413         struct xenbus_device *dev = be->dev;
 414         struct xenvif *vif;
 415
 416         if (be->vif != NULL)
 417                 return 0;
 418
 419         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 420         if (err != 1) {
 421                 xenbus_dev_fatal(dev, err, "reading handle");
 422                 return (err < 0) ? err : -EINVAL;
 423         }
 424
 425         vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 426         if (IS_ERR(vif)) {
 427                 err = PTR_ERR(vif);
 428                 xenbus_dev_fatal(dev, err, "creating interface");
 429                 return err;
 430         }
 431         be->vif = vif;
 432
 433         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 434         return 0;
 435 }
 436
 437 static void backend_disconnect(struct backend_info *be)
 438 {
 439         if (be->vif) {
 440                 xen_unregister_watchers(be->vif);
 441 #ifdef CONFIG_DEBUG_FS
 442                 xenvif_debugfs_delif(be->vif);
 443 #endif /* CONFIG_DEBUG_FS */
 444                 xenvif_disconnect(be->vif);
 445         }
 446 }
 447
 448 static void backend_connect(struct backend_info *be)
 449 {
 450         if (be->vif)
 451                 connect(be);
 452 }
 453
 454 static inline void backend_switch_state(struct backend_info *be,
 455                                         enum xenbus_state state)
 456 {
 457         struct xenbus_device *dev = be->dev;
 458
 459         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 460         be->state = state;
 461
 462         /* If we are waiting for a hotplug script then defer the
 463          * actual xenbus state change.
 464          */
 465         if (!be->have_hotplug_status_watch)
 466                 xenbus_switch_state(dev, state);
 467 }
 468
 469 /* Handle backend state transitions:
 470  *
 471  * The backend state starts in InitWait and the following transitions are
 472  * allowed.
 473  *
 474  * InitWait -> Connected
 475  *
 476  *    ^    \         |
 477  *    |     \        |
 478  *    |      \       |
 479  *    |       \      |
 480  *    |        \     |
 481  *    |         \    |
 482  *    |          V   V
 483  *
 484  *  Closed  <-> Closing
 485  *
 486  * The state argument specifies the eventual state of the backend and the
 487  * function transitions to that state via the shortest path.
 488  */
 489 static void set_backend_state(struct backend_info *be,
 490                               enum xenbus_state state)
 491 {
 492         while (be->state != state) {
 493                 switch (be->state) {
 494                 case XenbusStateClosed:
 495                         switch (state) {
 496                         case XenbusStateInitWait:
 497                         case XenbusStateConnected:
 498                                 pr_info("%s: prepare for reconnect\n",
 499                                         be->dev->nodename);
 500                                 backend_switch_state(be, XenbusStateInitWait);
 501                                 break;
 502                         case XenbusStateClosing:
 503                                 backend_switch_state(be, XenbusStateClosing);
 504                                 break;
 505                         default:
 506                                 BUG();
 507                         }
 508                         break;
 509                 case XenbusStateInitWait:
 510                         switch (state) {
 511                         case XenbusStateConnected:
 512                                 backend_connect(be);
 513                                 backend_switch_state(be, XenbusStateConnected);
 514                                 break;
 515                         case XenbusStateClosing:
 516                         case XenbusStateClosed:
 517                                 backend_switch_state(be, XenbusStateClosing);
 518                                 break;
 519                         default:
 520                                 BUG();
 521                         }
 522                         break;
 523                 case XenbusStateConnected:
 524                         switch (state) {
 525                         case XenbusStateInitWait:
 526                         case XenbusStateClosing:
 527                         case XenbusStateClosed:
 528                                 backend_disconnect(be);
 529                                 backend_switch_state(be, XenbusStateClosing);
 530                                 break;
 531                         default:
 532                                 BUG();
 533                         }
 534                         break;
 535                 case XenbusStateClosing:
 536                         switch (state) {
 537                         case XenbusStateInitWait:
 538                         case XenbusStateConnected:
 539                         case XenbusStateClosed:
 540                                 backend_switch_state(be, XenbusStateClosed);
 541                                 break;
 542                         default:
 543                                 BUG();
 544                         }
 545                         break;
 546                 default:
 547                         BUG();
 548                 }
 549         }
 550 }
 551
 552 /**
 553  * Callback received when the frontend's state changes.
 554  */
 555 static void frontend_changed(struct xenbus_device *dev,
 556                              enum xenbus_state frontend_state)
 557 {
 558         struct backend_info *be = dev_get_drvdata(&dev->dev);
 559
 560         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 561
 562         be->frontend_state = frontend_state;
 563
 564         switch (frontend_state) {
 565         case XenbusStateInitialising:
 566                 set_backend_state(be, XenbusStateInitWait);
 567                 break;
 568
 569         case XenbusStateInitialised:
 570                 break;
 571
 572         case XenbusStateConnected:
 573                 set_backend_state(be, XenbusStateConnected);
 574                 break;
 575
 576         case XenbusStateClosing:
 577                 set_backend_state(be, XenbusStateClosing);
 578                 break;
 579
 580         case XenbusStateClosed:
 581                 set_backend_state(be, XenbusStateClosed);
 582                 if (xenbus_dev_is_online(dev))
 583                         break;
 584                 /* fall through if not online */
 585         case XenbusStateUnknown:
 586                 set_backend_state(be, XenbusStateClosed);
 587                 device_unregister(&dev->dev);
 588                 break;
 589
 590         default:
 591                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 592                                  frontend_state);
 593                 break;
 594         }
 595 }
 596
 597
 598 static void xen_net_read_rate(struct xenbus_device *dev,
 599                               unsigned long *bytes, unsigned long *usec)
 600 {
 601         char *s, *e;
 602         unsigned long b, u;
 603         char *ratestr;
 604
 605         /* Default to unlimited bandwidth. */
 606         *bytes = ~0UL;
 607         *usec = 0;
 608
 609         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 610         if (IS_ERR(ratestr))
 611                 return;
 612
 613         s = ratestr;
 614         b = simple_strtoul(s, &e, 10);
 615         if ((s == e) || (*e != ','))
 616                 goto fail;
 617
 618         s = e + 1;
 619         u = simple_strtoul(s, &e, 10);
 620         if ((s == e) || (*e != '\0'))
 621                 goto fail;
 622
 623         *bytes = b;
 624         *usec = u;
 625
 626         kfree(ratestr);
 627         return;
 628
 629  fail:
 630         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 631         kfree(ratestr);
 632 }
 633
 634 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 635 {
 636         char *s, *e, *macstr;
 637         int i;
 638
 639         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 640         if (IS_ERR(macstr))
 641                 return PTR_ERR(macstr);
 642
 643         for (i = 0; i < ETH_ALEN; i++) {
 644                 mac[i] = simple_strtoul(s, &e, 16);
 645                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 646                         kfree(macstr);
 647                         return -ENOENT;
 648                 }
 649                 s = e+1;
 650         }
 651
 652         kfree(macstr);
 653         return 0;
 654 }
 655
 656 static void xen_net_rate_changed(struct xenbus_watch *watch,
 657                                 const char **vec, unsigned int len)
 658 {
 659         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 660         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 661         unsigned long   credit_bytes;
 662         unsigned long   credit_usec;
 663         unsigned int queue_index;
 664
 665         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 666         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 667                 struct xenvif_queue *queue = &vif->queues[queue_index];
 668
 669                 queue->credit_bytes = credit_bytes;
 670                 queue->credit_usec = credit_usec;
 671                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 672                         queue->remaining_credit > queue->credit_bytes) {
 673                         queue->remaining_credit = queue->credit_bytes;
 674                 }
 675         }
 676 }
 677
 678 static int xen_register_watchers(struct xenbus_device *dev, struct xenvif *vif)
 679 {
 680         int err = 0;
 681         char *node;
 682         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 683
 684         if (vif->credit_watch.node)
 685                 return -EADDRINUSE;
 686
 687         node = kmalloc(maxlen, GFP_KERNEL);
 688         if (!node)
 689                 return -ENOMEM;
 690         snprintf(node, maxlen, "%s/rate", dev->nodename);
 691         vif->credit_watch.node = node;
 692         vif->credit_watch.callback = xen_net_rate_changed;
 693         err = register_xenbus_watch(&vif->credit_watch);
 694         if (err) {
 695                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 696                 kfree(node);
 697                 vif->credit_watch.node = NULL;
 698                 vif->credit_watch.callback = NULL;
 699         }
 700         return err;
 701 }
 702
 703 static void xen_unregister_watchers(struct xenvif *vif)
 704 {
 705         if (vif->credit_watch.node) {
 706                 unregister_xenbus_watch(&vif->credit_watch);
 707                 kfree(vif->credit_watch.node);
 708                 vif->credit_watch.node = NULL;
 709         }
 710 }
 711
 712 static void unregister_hotplug_status_watch(struct backend_info *be)
 713 {
 714         if (be->have_hotplug_status_watch) {
 715                 unregister_xenbus_watch(&be->hotplug_status_watch);
 716                 kfree(be->hotplug_status_watch.node);
 717         }
 718         be->have_hotplug_status_watch = 0;
 719 }
 720
 721 static void hotplug_status_changed(struct xenbus_watch *watch,
 722                                    const char **vec,
 723                                    unsigned int vec_size)
 724 {
 725         struct backend_info *be = container_of(watch,
 726                                                struct backend_info,
 727                                                hotplug_status_watch);
 728         char *str;
 729         unsigned int len;
 730
 731         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 732         if (IS_ERR(str))
 733                 return;
 734         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 735                 /* Complete any pending state change */
 736                 xenbus_switch_state(be->dev, be->state);
 737
 738                 /* Not interested in this watch anymore. */
 739                 unregister_hotplug_status_watch(be);
 740         }
 741         kfree(str);
 742 }
 743
 744 static void connect(struct backend_info *be)
 745 {
 746         int err;
 747         struct xenbus_device *dev = be->dev;
 748         unsigned long credit_bytes, credit_usec;
 749         unsigned int queue_index;
 750         unsigned int requested_num_queues;
 751         struct xenvif_queue *queue;
 752
 753         /* Check whether the frontend requested multiple queues
 754          * and read the number requested.
 755          */
 756         err = xenbus_scanf(XBT_NIL, dev->otherend,
 757                            "multi-queue-num-queues",
 758                            "%u", &requested_num_queues);
 759         if (err < 0) {
 760                 requested_num_queues = 1; /* Fall back to single queue */
 761         } else if (requested_num_queues > xenvif_max_queues) {
 762                 /* buggy or malicious guest */
 763                 xenbus_dev_fatal(dev, err,
 764                                  "guest requested %u queues, exceeding the maximum of %u.",
 765                                  requested_num_queues, xenvif_max_queues);
 766                 return;
 767         }
 768
 769         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 770         if (err) {
 771                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 772                 return;
 773         }
 774
 775         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 776         xen_unregister_watchers(be->vif);
 777         xen_register_watchers(dev, be->vif);
 778         read_xenbus_vif_flags(be);
 779
 780         /* Use the number of queues requested by the frontend */
 781         be->vif->queues = vzalloc(requested_num_queues *
 782                                   sizeof(struct xenvif_queue));
 783         be->vif->num_queues = requested_num_queues;
 784         be->vif->stalled_queues = requested_num_queues;
 785
 786         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 787                 queue = &be->vif->queues[queue_index];
 788                 queue->vif = be->vif;
 789                 queue->id = queue_index;
 790                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 791                                 be->vif->dev->name, queue->id);
 792
 793                 err = xenvif_init_queue(queue);
 794                 if (err) {
 795                         /* xenvif_init_queue() cleans up after itself on
 796                          * failure, but we need to clean up any previously
 797                          * initialised queues. Set num_queues to i so that
 798                          * earlier queues can be destroyed using the regular
 799                          * disconnect logic.
 800                          */
 801                         be->vif->num_queues = queue_index;
 802                         goto err;
 803                 }
 804
 805                 queue->credit_bytes = credit_bytes;
 806                 queue->remaining_credit = credit_bytes;
 807                 queue->credit_usec = credit_usec;
 808
 809                 err = connect_rings(be, queue);
 810                 if (err) {
 811                         /* connect_rings() cleans up after itself on failure,
 812                          * but we need to clean up after xenvif_init_queue() here,
 813                          * and also clean up any previously initialised queues.
 814                          */
 815                         xenvif_deinit_queue(queue);
 816                         be->vif->num_queues = queue_index;
 817                         goto err;
 818                 }
 819         }
 820
 821 #ifdef CONFIG_DEBUG_FS
 822         xenvif_debugfs_addif(be->vif);
 823 #endif /* CONFIG_DEBUG_FS */
 824
 825         /* Initialisation completed, tell core driver the number of
 826          * active queues.
 827          */
 828         rtnl_lock();
 829         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 830         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 831         rtnl_unlock();
 832
 833         xenvif_carrier_on(be->vif);
 834
 835         unregister_hotplug_status_watch(be);
 836         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 837                                    hotplug_status_changed,
 838                                    "%s/%s", dev->nodename, "hotplug-status");
 839         if (!err)
 840                 be->have_hotplug_status_watch = 1;
 841
 842         netif_tx_wake_all_queues(be->vif->dev);
 843
 844         return;
 845
 846 err:
 847         if (be->vif->num_queues > 0)
 848                 xenvif_disconnect(be->vif); /* Clean up existing queues */
 849         vfree(be->vif->queues);
 850         be->vif->queues = NULL;
 851         be->vif->num_queues = 0;
 852         return;
 853 }
 854
 855
 856 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue)
 857 {
 858         struct xenbus_device *dev = be->dev;
 859         unsigned int num_queues = queue->vif->num_queues;
 860         unsigned long tx_ring_ref, rx_ring_ref;
 861         unsigned int tx_evtchn, rx_evtchn;
 862         int err;
 863         char *xspath;
 864         size_t xspathsize;
 865         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 866
 867         /* If the frontend requested 1 queue, or we have fallen back
 868          * to single queue due to lack of frontend support for multi-
 869          * queue, expect the remaining XenStore keys in the toplevel
 870          * directory. Otherwise, expect them in a subdirectory called
 871          * queue-N.
 872          */
 873         if (num_queues == 1) {
 874                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 875                 if (!xspath) {
 876                         xenbus_dev_fatal(dev, -ENOMEM,
 877                                          "reading ring references");
 878                         return -ENOMEM;
 879                 }
 880                 strcpy(xspath, dev->otherend);
 881         } else {
 882                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 883                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 884                 if (!xspath) {
 885                         xenbus_dev_fatal(dev, -ENOMEM,
 886                                          "reading ring references");
 887                         return -ENOMEM;
 888                 }
 889                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 890                          queue->id);
 891         }
 892
 893         err = xenbus_gather(XBT_NIL, xspath,
 894                             "tx-ring-ref", "%lu", &tx_ring_ref,
 895                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 896         if (err) {
 897                 xenbus_dev_fatal(dev, err,
 898                                  "reading %s/ring-ref",
 899                                  xspath);
 900                 goto err;
 901         }
 902
 903         /* Try split event channels first, then single event channel. */
 904         err = xenbus_gather(XBT_NIL, xspath,
 905                             "event-channel-tx", "%u", &tx_evtchn,
 906                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 907         if (err < 0) {
 908                 err = xenbus_scanf(XBT_NIL, xspath,
 909                                    "event-channel", "%u", &tx_evtchn);
 910                 if (err < 0) {
 911                         xenbus_dev_fatal(dev, err,
 912                                          "reading %s/event-channel(-tx/rx)",
 913                                          xspath);
 914                         goto err;
 915                 }
 916                 rx_evtchn = tx_evtchn;
 917         }
 918
 919         /* Map the shared frame, irq etc. */
 920         err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref,
 921                              tx_evtchn, rx_evtchn);
 922         if (err) {
 923                 xenbus_dev_fatal(dev, err,
 924                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
 925                                  tx_ring_ref, rx_ring_ref,
 926                                  tx_evtchn, rx_evtchn);
 927                 goto err;
 928         }
 929
 930         err = 0;
 931 err: /* Regular return falls through with err == 0 */
 932         kfree(xspath);
 933         return err;
 934 }
 935
 936 static int read_xenbus_vif_flags(struct backend_info *be)
 937 {
 938         struct xenvif *vif = be->vif;
 939         struct xenbus_device *dev = be->dev;
 940         unsigned int rx_copy;
 941         int err, val;
 942
 943         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
 944                            &rx_copy);
 945         if (err == -ENOENT) {
 946                 err = 0;
 947                 rx_copy = 0;
 948         }
 949         if (err < 0) {
 950                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
 951                                  dev->otherend);
 952                 return err;
 953         }
 954         if (!rx_copy)
 955                 return -EOPNOTSUPP;
 956
 957         if (xenbus_scanf(XBT_NIL, dev->otherend,
 958                          "feature-rx-notify", "%d", &val) < 0)
 959                 val = 0;
 960         if (!val) {
 961                 /* - Reduce drain timeout to poll more frequently for
 962                  *   Rx requests.
 963                  * - Disable Rx stall detection.
 964                  */
 965                 be->vif->drain_timeout = msecs_to_jiffies(30);
 966                 be->vif->stall_timeout = 0;
 967         }
 968
 969         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
 970                          "%d", &val) < 0)
 971                 val = 0;
 972         vif->can_sg = !!val;
 973
 974         vif->gso_mask = 0;
 975         vif->gso_prefix_mask = 0;
 976
 977         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
 978                          "%d", &val) < 0)
 979                 val = 0;
 980         if (val)
 981                 vif->gso_mask |= GSO_BIT(TCPV4);
 982
 983         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
 984                          "%d", &val) < 0)
 985                 val = 0;
 986         if (val)
 987                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
 988
 989         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
 990                          "%d", &val) < 0)
 991                 val = 0;
 992         if (val)
 993                 vif->gso_mask |= GSO_BIT(TCPV6);
 994
 995         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
 996                          "%d", &val) < 0)
 997                 val = 0;
 998         if (val)
 999                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
1000
1001         if (vif->gso_mask & vif->gso_prefix_mask) {
1002                 xenbus_dev_fatal(dev, err,
1003                                  "%s: gso and gso prefix flags are not "
1004                                  "mutually exclusive",
1005                                  dev->otherend);
1006                 return -EOPNOTSUPP;
1007         }
1008
1009         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
1010                          "%d", &val) < 0)
1011                 val = 0;
1012         vif->ip_csum = !val;
1013
1014         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
1015                          "%d", &val) < 0)
1016                 val = 0;
1017         vif->ipv6_csum = !!val;
1018
1019         return 0;
1020 }
1021
1022 static const struct xenbus_device_id netback_ids[] = {
1023         { "vif" },
1024         { "" }
1025 };
1026
1027 static struct xenbus_driver netback_driver = {
1028         .ids = netback_ids,
1029         .probe = netback_probe,
1030         .remove = netback_remove,
1031         .uevent = netback_uevent,
1032         .otherend_changed = frontend_changed,
1033 };
1034
1035 int xenvif_xenbus_init(void)
1036 {
1037         return xenbus_register_backend(&netback_driver);
1038 }
1039
1040 void xenvif_xenbus_fini(void)
1041 {
1042         return xenbus_unregister_driver(&netback_driver);
1043 }